Skip to content

Integration patterns

Plasma is designed to exchange data with the outside world — and to move it between layers internally — through a small set of well-defined patterns. The backbone is ECST (Event-Carried State Transfer): components publish state changes as events, and consumers react without querying back.

Pub/Sub messaging

The Integration layer is the platform's central nervous system, built on NATS as the event bus. Every layer runs its own bus and exchanges data through explicit relays, so there is no central coordinator — behavior emerges from the channel topology.

Layer bus Technology Carries
Integration NATS business events (ECST)
Cognition Kafka analytical data streams
Conversation Matrix / Synapse natural-language utterances

Connectors

Two complementary tools bring external data in and move it around:

  • Airbyte — the connect application ingests from external systems (databases, SaaS APIs, files) using Airbyte's connector catalog, with a custom Plasma destination that lands data directly on the platform's bus.
  • Benthos — lightweight stream processors used as relays between layers, providing backpressure so a fast producer can't overwhelm a slower consumer.

Workflow automation (n8n) and durable orchestration (Temporal) are available in the Interaction layer for connecting steps across systems.

Unified query

Trino provides a single SQL surface across all of a platform's data sources, so you can query the platform as one logical database regardless of where each dataset physically lives. The MCP server exposes these queries to AI assistants over SSE.

APIs & contracts

Plasma treats its ontology as the contract. Entities and metrics are defined as Protobuf schemas — the .proto file is the source of truth:

message Person {
  string id = 1 [(is_primary_key)=true];
  string name = 2;
}

From those definitions the platform generates JSON Schema and registers it in a runtime schema registry (Apicurio), so producers and consumers validate against the same, versioned contract.

→ For driving plasmactl itself from scripts and CI, see Automation & structured output.